Every so often, we find the most interesting data science links from around the web and collect them in Data Science Briefings, the DataMiningApps newsletter. Subscribe now for free if you want to be the first to get up to speed on interesting resources.
Stable Diffusion 2.0 is here!
- Stable Diffusion 2.0 Release
Stable Diffusion 2.0 delivers a number of big improvements and features versus the original V1 release. Also check out this web UI. - Full Pose Control For Generative Art
In this article we focus on a new feature called depth2img. This very powerful new feature lets you create new images from reference pictures and it opens up new ways to control pose on an image generated by Stable Diffusion. - Latent Video Diffusion Models for High-Fidelity Video Generation with Arbitrary Lengths
“Extensive experiments on various datasets and generated lengths suggest that our framework is able to sample much more realistic and longer videos than previous approaches, including GAN-based, autoregressive-based, and diffusion-based methods.” - Magic3D: High-Resolution Text-to-3D Content Creation
“Magic3D is a new text-to-3D content creation tool that creates 3D mesh models with unprecedented quality.” - MAGIC VIDEO
Efficient Video Generation With Latent Diffusion Models - SinDiffusion: Learning a Diffusion Model from a Single Natural Image
“This repository implements the SinDiffusion model, leveraging denoising diffusion models to capture internal distribution of patches from a single natural image.”
Facebook releases Galactica and quickly decomissions it
- Meta AI set out on a mission to organise scientific knowledge with a large language model. They used 48 million papers and text books and published an online demo called Galactica.
“You can use it to write scientific papers & code, summarise academic literature, solve math problems.” - Soon enough, Twitter was full of negative examples of Galactica’s hallucinations
… showing many falsehoods, racist, wrong, totally inaccurate or invented theorems or statements. - A few days later, Meta decided to take down Galactica. Gary Marcus, a researcher who’s been very critical about the implications of these kind of hallucinations, posted a video on the limitations of DL and LLMs
- Also, AI Center, Uni of Hong-Kong published a Survey of Hallucination in Natural Language Generation in which they provide a taxonomy of LLMs hallucinations and ideas for remediating this issue.
- In a paper published this week titled The CRINGE Loss (yes, really), another team at Meta has come up with a new learning approach for training the LLM on what it should not do
An issue with LLMs is that they’re pre-trained with human baseline documents, treating all training data as positive instances. - CohereAI wrote about 5 Ways to Tackle the Challenges of Large Language Models.
- The Stanford Institute for Human-Centered AI is doing intensive research on the capabilities, limitations, and risks of LLMs
Just this week, they published Holistic Evaluation of Language Models in which they define a taxonomy of LLMs scenarios, and a series of metrics and benchmark to better evaluate and understand LLMs.
Still, large language models continue to grow
- Accelerating Document AI
“Document AI includes many data science tasks from image classification, image to text, document question answering, table question answering, and visual question answering.” - Document AI: LiLT a better language agnostic LayoutLM model
“In this blog, you will learn how to fine-tune LiLt for document-understand using Hugging Face Transformers.”
And more
- An offshore workforce is training Amazon’s warehouse-monitoring algorithms
The company pays workers in India and Costa Rica to review warehouse camera footage for just hundreds of dollars a month. - Intel’s new deepfake detector can spot a real or fake video based on blood flow in video pixels
Intel’s FakeCatcher detects a deepfake in real time with a 96% accuracy rate - How Uber Optimizes the Timing of Push Notifications using ML and Linear Programming
We introduced a system we call the Consumer Communication Gateway (CCG): a centralized intelligence layer to manage the quality, ranking, timing, and frequency of push notifications on a user level. - How AI Text Generation Models Are Reshaping Customer Support at Airbnb
Leveraging text generation models to build more effective, scalable customer support products. - Art Made With Artificial Intelligence Wins at State Fair
Artist Jason Allen placed first in a Colorado contest, generating debate about A.I.’s role in art - Who ranks better? Memgraph vs NetworkX PageRank
Memgraph is an open-source in-memory graph computation platform for static or real-time graph analytics. - Who needs MLflow when you have SQLite?
“I spent about six years working as a data scientist and tried to use MLflow several times (and others as well) to track my experiments; however, every time I tried using it, I abandoned it a few days after.” - Challenges of Building Realtime ML Pipelines
“Realtime machine learning is on the rise, and as companies start introducing realtime into their ML pipelines, they are finding themselves having to weigh the trade-offs between performance, cost, and infrastructure complexity, and determine which to prioritize.” - Universality of Neural Networks on Graphs vs. Sets
You can think of sets as graphs with just nodes and no edges. - Announcing the NeurIPS 2022 Awards
The award-winning papers for NeurIPS 2022 are here! - rtdl
Research on Tabular Deep Learning. - Scikit-learn with Transformers (with skops)
skops is a Python library helping you share your scikit-learn based models and put them in production. - Kubeshark, the API Traffic Viewer for kubernetes
Provides deep visibility and monitoring of all API traffic and payloads going in, out and across containers and pods inside a Kubernetes cluster - Playtesting Candycrush
“Today I learned that there’s actual research in playtesting video games using deep learning.”
